The public service’s AI Review Committee has met. Its terms of reference send the advice to the minister and a summary to the public

In August we looked at the AI Review Committee, the body the APS AI Plan created to advise on the riskiest government uses of AI, and ended on one open question: nothing published said whether its advice would become public. An archived copy of the committee’s page from 20 August carried members and a description, but no terms of reference and no communique.

Both now exist. On 7 October the DTA said the committee has met, and the committee page now says its terms of reference were presented at the inaugural meeting on 26 June 2026 and endorsed by members. They answer the August question directly: under the terms, the advice goes to the agency and the minister, and the public gets a summary.

The first two cases were for staff

The communique describes both use cases in a few sentences each, and both are tools for the agencies’ own staff. The Attorney-General’s Department discussed the potential to use AI to support staff productivity and wellbeing by assisting with document analysis, matter triage and workflow optimisation. Services Australia presented a proof of concept to help staff search its Operational Blueprint, a library of operational instructions for staff, tested with a small cohort of experienced users.

The two use cases at the committee’s 26 June 2026 meeting, as the communique describes them. Quoted words are the communique’s. Source: AI Review Committee Communique, 26 June 2026, pages 1 and 2.

The communique does not say which of the committee’s triggers, a residual high-risk rating or a use case an agency judges highly sensitive, novel or complex, brought either case to it. The DTA’s release says only that the two were referred for consideration. The communique’s own framing is that the committee looks at use cases from early conceptual thinking about potential capabilities through to testing capabilities in a pilot, and these two sit at those two ends.

Who gets what

The terms of reference are five pages and plain. The committee is accountable to the Minister for Finance. After a use case is considered, the secretariat drafts the advice, which must be approved by the Chair before being provided to the agency. Then: Following each meeting, the DTA will provide the minister with meeting minutes, and the advice issued to agencies. The public’s share is in section 4: the terms of reference and membership are published, and The Committee will publish communiqués on meeting outcomes on digital.gov.au.

Clause numbers are from the AI Review Committee Terms of reference (dated July 2026). “No provision” means the five pages do not provide for that recipient; it does not mean the record could never be released some other way. A communique and the terms are public, so everyone has them.

The first communique is more specific than a press release. It lists what members encouraged each agency to work on, and the Services Australia note reads as a critique of a trial measured on user satisfaction. It is also expressly a summary of key discussion themes. What the agencies were actually told stays with them and the minister.

What the advice is not

The advice provided by the Committee will not … constitute or be taken as any form of endorsement, sanction, immunity or compliance advice with respect to any AI use case

AI Review Committee Terms of reference, clause 2.9

That clause matters because of who sits on the committee. Its page lists six members: the DTA’s Chief Executive Officer as Chair, the General Manager of the AI Safety Institute, the Privacy Commissioner, the National Cyber Security Coordinator, and senior officials from the Australian Public Service Commission and the Australian Taxation Office. The terms anticipate the obvious worry about a regulator in that room: Members with regulatory functions are not constrained in exercising those functions with respect to any act, practice or agency which is subject to review by the Committee. A member also recuses where the member’s home agency presents a use case for review.

Three other clauses shape how much weight a single meeting carries. The advice reflects the consensus position of its members. A quorum is four appointed members, but If a quorum is not met, a meeting may still proceed at the discretion of the Chair. And the committee convenes meetings on an as-needed basis. As at 7 October, its communiques page lists one meeting.

Measured against the plan’s own milestones

The APS AI Plan’s Appendix A gave the committee three dated stages, and the first included a step that is easy to check from outside.

1. Mid 2026Plan: establish. Identify membership, establish ToRs and secretariat, and amend policy and AI impact tool to reflect new high-risk use case workflow
2. 26 Jun 2026Inaugural meeting; terms of reference endorsed; two use cases considered
3. 20 Aug 2026Archived committee page: members listed, no terms of reference or communique
4. 7 Oct 2026DTA release; terms of reference and first communique on the committee page
5. Late 2026Plan: operationalise. Initial review of 1-2 high risk AI projects. Provision of advice, and reflection of lessons learned
6. Late 2026 to early 2027Plan: full maturity. review of all high risk cases, with meetings every 6 weeks, written advice to agencies and full integration into impact assessment workflows

Plan stages and their wording from the APS AI Plan 2025, Appendix A, AI Review Committee row; meeting date from the communique; 20 August from the Internet Archive copy of the committee page; 7 October from the DTA release. Read 7 October 2026.

Membership and terms are published, and the terms refer to a secretariat. The policy and tool amendment is not visible in what is published. On 7 October the Policy for the responsible use of AI in government on digital.gov.au was still version 2.0, effective 15 December 2025, and its 22-page PDF does not contain the word committee. Its high-risk steps tell an agency to report the use case to its accountable official, govern it through a designated board or senior executive, report it to the DTA once deployed and review it at least every 12 months. None of them mentions the committee. The impact assessment tool, its download marked last updated 1 December 2025, and the tool’s guidance PDF do not mention it either.

The committee’s own terms describe what it reviews as use cases that an agency has either rated residual high-risk or identified as highly sensitive, novel or complex. Until the policy is amended, how a case reaches the committee is not set out in the policy that binds agencies.

Our view

This is a solid first showing. The committee met when the plan said it would, published real terms, and wrote a communique that names agencies and says what it pressed them on. The clause that keeps regulators free to act is the right one to have written down.

But the arrangement answers our August question the cautious way. Advice on the government’s riskiest AI uses goes to the agency and the minister; the public gets a summary the Chair approves. That is defensible for an internal advisory body. It means the test we named then, an agency proceeding after this committee advises against it, would reach the public only if the communique chose to say so. The first two cases, an early idea and a trial, both for staff, did not come near that test. The plan’s next milestone, one or two high risk projects, should. When it arrives, the policy should send cases to the committee in its own words, and the communique on a high risk case should say plainly whether the agency went ahead.

Sources

1. Digital Transformation Agency, Inaugural AI Review Committee sets out expert advice on public sector AI use, media release, 7 October 2026 (read in full 7 October 2026): the committee has met; the 26 June meeting; two use cases referred for consideration.
2. Digital Transformation Agency, AI Review Committee, digital.gov.au (read in full 7 October 2026): the six members and their roles; terms of reference presented and endorsed at the 26 June 2026 meeting; the commitment to publish a communique after each meeting.
3. Digital Transformation Agency, Artificial Intelligence Review Committee: Terms of reference, dated July 2026, 5 pages (PDF read in full 7 October 2026): clauses 2.2, 2.8, 2.9 (p 2), 3.1 and 3.3 (p 2), 4.1 and 4.2, 6.2 (p 3), 7.2 (p 4), 9.2, 9.3 and 9.5 (p 5).
4. Digital Transformation Agency, Artificial Intelligence Review Committee Communique, dated 26 June 2026, 2 pages (PDF read in full 7 October 2026): the two use cases and what members encouraged; not an endorsement or approval.
5. Digital Transformation Agency, Communiques, digital.gov.au (read 7 October 2026): one meeting listed, 26 June 2026.
6. Digital Transformation Agency, AI Plan for the Australian Public Service 2025, Appendix A: Plan deliverables (read 7 October 2026): the AI Review Committee row, its three stages and timeframes.
7. Digital Transformation Agency, Policy for the responsible use of AI in government, version 2.0, 22 pages (PDF full-text searched 7 October 2026), with its web pages Implementation (effective 15 December 2025) and AI use case impact assessment (the high-risk steps).
8. Digital Transformation Agency, AI impact assessment tool, introduction page with the tool DOCX (marked last updated 1 December 2025) and the guidance PDF (both full-text searched 7 October 2026).
9. Internet Archive copy of the AI Review Committee page, 20 August 2026 (read 7 October 2026): no terms of reference, meeting or communique on the page.

Read the terms of reference differently, or know of a revised policy? Tell us and we will check it against the documents and log the outcome here.